Use of cluster analysis for the study of machine processes

The study of machine processes using the mean-tracking cluster algorithm has provided useful results and a greater understanding of the processes. Clusters denoting faulty and fault-free machine behaviour derived from analyses in Sutanto and Warwick (1995) have provided guidance as to which regions to operate in and which to avoid. However, some clusters could not be so easily defined since they contained overlapping regions of faulty and fault-free machine behaviour. Clustering on process trajectories has separated these regions to a degree that they could be better defined. In doing so, correlations between errors were made more visible. The clustering exercise was then taken further to identify regions of successful and troublesome start-ups by clustering on start-up data. The results did not only define these regions successfully, but revealed that each start-up that was carried out had some probability of success and was never a non-starter start-up. (5 pages)